Methodology
MethodologyHow we measure every number
Every figure on this site uses the same definitions. This page explains what we count, how we count it, and what we leave out.
Counting rule
The rule that comes before every percentage
- An object counts as automated only if it needs no manual edit.
- A one-line fix counts as manual.
- Dead code is excluded from the denominator and reported separately.
- Objects resolved by an approved resolver proposal without further edit are reported as assisted, not automated.
Principles
How does Novuz publish a number?
- A number comes from delivered estates, not a lab test or a forecast.
- We publish figures in their typical form until each customer gives written permission to name them and their numbers.
- We use the same definitions on every engagement.
- We show the scope, so you can compare it with your own estate.
- If a customer asks us not to publish, we leave the number out.
Definitions
The four definitions
- Object: a unit of convertible logic (stored procedure, function, view, trigger, SSIS package, BTEQ script, Hive query or UDF, Snowflake procedure or task, job definition); counted once, dead objects reported separately.
- Automation rate: share of objects converted with no manual edit.
- Reconciliation accuracy: share of objects whose converted output passed schema, row-count and column-level checks against the original on identical inputs.
- Elapsed time: code received to reconciled output, in business days, excluding time waiting on the customer.
Reconciliation
How each object is checked
We run the original object and the converted object on the same inputs, then compare the outputs. Every object gets its own result in the reconciliation report: pass, fail or review.
- Schema: the same columns, types and keys.
- Row counts, per output table.
- Aggregates: sums, counts and distinct counts on every numeric and key column.
- Sampled rows, matched and compared field by field.
- Business-logic checks supplied by your own subject experts.
- Floating-point values match within a stated tolerance, listed in the report.
- Ordering is ignored unless the original code guarantees an order.
- Inputs are production data, or a masked or sampled copy chosen by you.
- Every miss is listed by line, with the cause and the fix.
The resolver
Where a language model is used, and where it is not
The resolver uses a language model for the roughly 5% of objects that fixed rules do not cover, on a model you approve, at 700 to 1,000 tokens per edge case, and your engineer approves every proposal. Nothing deploys unapproved.
Cost figures
How are prices and savings stated?
- Prices are in US dollars and fixed per engagement, as listed on the Pricing page.
- Cost comparisons are against a systems integrator’s hours-based rewrite of the same estate: about 40% cheaper, at a fixed price set before work starts.
- We publish no percentage against LLM converters, because none has completed a full migration to compare against. The honest contrast is a fixed price against open-ended token pricing.
- Token figures are typical across engagements and vary by estate.
- Charts marked “Illustrative” show a typical pattern, not one customer’s data.
Check it yourself
How to test our numbers on your own code
- Convert the same objects twice and diff the output. The diff should be empty.
- Reconcile on your own inputs, not ours.
- Read the resolver proposals and the engineer-approval log.
- Change one rule, re-run, and check that only the objects using that pattern change.
- Run your own linter over the converted code.
Limits
What do these numbers not tell you?
Results depend on the estate. Heavy dynamic SQL, undocumented business rules or uncommon source features can lower the automation rate. The estate assessment gives you a number for your own code in three business days, before you commit.
FAQ
Frequently asked questions
Last reviewed:
Find out what your logic debt is costing you.
Every object counted, lineage, dead code and a fixed price in three business days. $3,000, credited.
Not sure which door? The scorecard takes ten minutes and tells you.